Over the last 30 years, computing architecture has swung back and forth between two extremes: centralization and decentralization. In the early 60s, 70s, and 80s, computing was mostly done on giant mainframes in large organizations. But the advent of personal computers and the “client–server” framework in the early 1980s meant that computing became decentralized. The smartphones and mobile applications of the early 2000s further accelerated this decentralization. Yet, over the last decade, cloud computing has reversed this process. Once again, computing is increasingly being centralized in mega datacenters, like the ones run by public cloud providers such as Amazon, Google, Alibaba, Microsoft, and IBM.
Cloud computing has already had a transformative effect. In our earlier blog, “Can Enterprise Datacenters Co-Exist with Cloud?” we discussed how factors like scalability, convenience, cost savings, capital efficiencies, and flexible business models will help grow the cloud computing market from $246B in 2017 to $383B by 2020. But does this impressive growth rate represent an inevitable march toward the centralization of storage and computing? The future will likely mirror rather than overturn the historic swings between centralization and decentralization. Why, you may ask? Because cloud computing is not without its challenges.
The first challenge is latency, which includes the total time required for data from a device sensor to be transmitted back to the cloud, data analysis on the cloud server, and the transmission of instructions back to the device. Latency is hard enough in the digital world—websites that take more than 0.5 seconds to load on your browser typically see about a 20% drop in traffic—but it is especially important in time-critical use cases. For example, in autonomous vehicles, digital voice assistants (see “How Do Digital Voice Assistants (e.g. Alexa, Siri) Work?”), remote surgery, artificial reality (AR), and virtual reality (VR), near-instantaneous analyses and decision-making are needed.
The second challenge is bandwidth. Even though network bandwidth has increased (see “Why is 5G Disruptive?”), the impending massive deployment of the Internet of Things (IoT) will start generating large amounts of data. According to a Cisco report, IoT devices will generate 500 zettabytes of data (1 zettabyte = 1,000,000,000,000,000,000,000 bytes) in 2019, and this growth will continue exponentially. Transmitting all this data directly to cloud servers for centralized computing will not only be architecturally inefficient, but also add additional data management costs to cloud and enterprise datacenters.
The third challenge to cloud computing has multiple aspects. Transferring all data to the cloud involves privacy, security, and regulatory challenges, especially when personally identifiable information (PII) is involved. It’s for this reason that Apple stores users’ fingerprints or facial recognition data inside a secure enclave on the user’s iPhone: “It’s never stored on Apple servers, it’s never backed up to iCloud or anywhere else, and it can’t be used to match against other fingerprint databases.”
Finally, cloud computing poses undefined operational availability challenges that can arise from a single point of failure. For example, during a mega-outage at Amazon Web Services (AWS) earlier this year, dozens of AWS-hosted consumer applications, including Expedia, GitHub, and Flipboard, were available only intermittently for over five hours. Most businesses want end-user devices and applications to be functionally active even when internet connectivity is intermittent.
So, how can all these challenges be overcome? Enter edge computing, which solves the above challenges by taking a different approach than cloud computing. In the latter, data is computed at one or many mega datacenters, but edge computing takes place at or near the data source. In other words, computing moves from the “center” (either the cloud or an enterprise datacenter) to the “edge” (e.g., smartphones, drones, autonomous cars, robots, or any other intelligent machine tethered to the Internet)—hence the name.
According to Gartner, today, only 10% of data is processed outside datacenters or the cloud, but that number is likely to reach 75% by 2022. By 2022, over 50% of companies are expected to spend more of their IT budget on storage, networks, and computing in edge locations than in their own datacenters, and the edge computing market is expected to grow annually at 35% to $6.7B. According to Michael Dell, “We think edge computing could be 100 times bigger than the Internet as we know it today. That may sound crazy right now but give it a few years and I think that will be more understood.”
Edge computing will continue to grow for a wide range of consumer and business use cases, leveraging its inherent advantages in latency, bandwidth usage, security, and privacy. Examples of these cases include autonomous vehicles, where contextual data needs to be collected and analyzed quickly; industrial automation, where sensors can quickly analyze and act on ambient signals; AR/VR devices, which can prevent motion sickness for users by minimizing data transmission latency; and sensor-based remote inspections in agriculture and industrials. As leading AI players bring machine learning capabilities to the edge, edge computing will receive an additional boost in growth. As an example, Google recently announced the launch of Edge TPU, which is an “ASIC chip running TensorFlow Lite ML [machine learning] models at the edge” that delivers “lightning fast ML inference at the edge. Your sensors become more than data collectors—they make local, real-time, intelligent decisions.”
So, what’s in store for cloud and edge computing? Well, there’s room for both. In our earlier blog, “Can Enterprise Datacenters Co-Exist with Cloud?” we concluded that the evolutionary path for enterprise datacenters and clouds is harmonious coexistence. Edge computing is also likely to coexist with public and private clouds. Depending on the specific requirements of use cases, such as latency, privacy, and bandwidth, businesses may decide to keep computing either in public/private clouds or at the edge.